About Yemen and a housewife

The data on our poster is presented visually, unlike reference tables which present territories and their sizes in a list. While sorting the territories by size based on Wikipedia, it became clear that areas of some territories do not exactly match their sizes.

One example is Yemen, which according to Wikipedia is larger than its neighboring territories that look larger than Yemen visually.

In our case everything is shown in one scale. Comparing territories in pixels without water gives a totally different rating.

  Wikipedia, km² Factbook, km² Pixels
1 Yemen 527 970 527 968 43 603
2 Thailand 511 944 510 890 48 322
3 Spain 499 532 498 980 46 953
4 Cameroon 469 261 472 710 45 466
5 Turkmenistan 467 131 469 930 44 986
6 Papua New Guinea 453 583 452 860 43 836

Ideally, any housewife can measure Yemen’s surface area on Google Maps with as much precision as she has time to spend.

Wikipedia claims Yemen has a surface area of 527 970 km² while the housewife was able to click up 454 716 km², which is 73 254 km² less.

Counting pixels for other territories.

After Yemen we see many more mismatches in the rating. So, what is really going on here?

To find out the truth, we need to decide on the authoritative source of information. Looking at Yandex, Google, Bing, Yahoo, OpenStreetMaps, ArcGIS and others as potential candidates for this role. All of them use the same WGS84 Web Mercator standard and the same coordinates, yet each of them has its own peculiarities.

First of all, we are interested in precise water boundaries. To choose the best source, we need to compare satellite images of some porous region, for example the area around Queen Maud Bay in Canada.

Yandex Google Bing Yahoo OSM ArcGIS Wikimapia

Google Maps turned out to be the most detailed.

Now we need to distinguish between pixels belonging to different territories. It’s an easy job for a human who can just look at territory borders, but we need to find a way to explain this to a calculator.

All online maps are made of tiles. A tile is a fragment made by 256×256 pixels. There are services that allow to customize Google’s tiles but unfortunately they don’t allow to change colors by territory. Another option, of course, is to color territories using polygons, but this method was abandoned almost immediately as it’s impossible to trace every single puddle of water this way.

Since we will need to color the territories by hand, we need to define the scale. There are 20 of them.

After some considerations it became clear that scale No. 10 is perfect for the job. To better imagine the size of this texture, look at the giant globe in the studio’s museum and a 30″ screen with resolution of 2600×1600 pixels.

Showing the entire Google Maps in scale 10 will require 4 000 of such displays.

Let’s assume that to color the entire map in scale 10 we will need one designer working full time for a month. Coloring it in scale 11 will require four designers, and scale 12 will need 16 of them. Based on that, it becomes evident that coloring territories in anything above scale 11 with the help of human designers is unfeasible.

Taking a close look at Vatican with an area of 0.5 km² (the world’s smallest territory).

In scale 10, Vatican will take up 11 pixels, which is about 20 pixels per 1 km2.

To finally put the issue of scale to rest, comparing St. Lucia in scale 12 with a pixel overlay created for the same territory in scale 10.

Customizing tiles so that only land and water pixels remain.

Stitching 262 144 customized tiles (each 256×256 pixels) into 64 giant megatiles with the resolution of 16 384×16 384 pixels. Each megatile is made of 4096 tiles.

Removing antialiasing to drastically simplify the process of counting and coloring pixels.

Drawing a matrix of 17×17 colors to color approximately 280 territories.

Marking the territories on a map, finding uncolored pixels, looking up which territory they belong to and giving them a unique color.

The matrix is filled.

To cut up the megatiles back into 262 144 parts, creating an improvised tile scanner in 3D Max.

Comparing the result with other maps in SAS.Planet.